-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathshow_tensorboard.py
186 lines (152 loc) · 9.27 KB
/
show_tensorboard.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
import os
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from tensorboard.backend.event_processing import event_accumulator
# 从TensorBoard日志目录加载事件文件
def load_tensorboard_data(log_dir):
ea = event_accumulator.EventAccumulator(log_dir)
ea.Reload() # 加载事件数据
return ea
# 计算给定数据的均值和方差
def calculate_mean_and_std(data):
mean = np.mean(data, axis=0)
std = np.std(data, axis=0)
return mean, std
# 提取指定tag的数据
def extract_data(ea, tag):
data = ea.Scalars(tag)
steps = [x.step for x in data]
values = [x.value for x in data]
return np.array(steps), np.array(values)
# 提取所有seed文件夹中的数据
def extract_seed_data(base_dir, tags):
all_seed_data = {}
# 遍历所有seed文件夹
for seed_folder in os.listdir(base_dir):
if seed_folder.endswith('_'):
continue
seed_path = os.path.join(base_dir, seed_folder)
if os.path.isdir(seed_path):
seed_data = {}
# 加载每个文件夹中的TensorBoard数据
for tag in tags:
ea = load_tensorboard_data(seed_path)
steps, values = extract_data(ea, tag)
if tag not in seed_data:
seed_data[tag] = []
seed_data[tag].append(values)
all_seed_data[seed_folder] = seed_data
return all_seed_data
# 对不同长度的数组进行对齐，使它们具有相同的形状
def align_data(all_seed_data, tag):
# 获取所有seeds的steps长度
seed_steps = [len(data[tag][0]) for data in all_seed_data.values()]
min_steps = min(seed_steps)
min_steps = min(min_steps, 100)
aligned_data = []
for seed_folder, seed_data in all_seed_data.items():
values = seed_data[tag]
for value in values:
if len(value) > min_steps:
aligned_value = value[:min_steps]
else:
aligned_value = value
aligned_data.append(aligned_value)
return np.array(aligned_data)
# 平滑函数：滑动平均或指数加权平均
def smooth_data(data, smooth_factor=0.8):
"""
对数据进行滑动平均或指数加权平均。
参数：
- data: 需要平滑的数据 (1D array)
- smooth_factor: 平滑因子，0-1之间，值越大越平滑
返回：
- 平滑后的数据
"""
smoothed = np.zeros_like(data)
smoothed[0] = data[0]
for t in range(1, len(data)):
smoothed[t] = smooth_factor * smoothed[t - 1] + (1 - smooth_factor) * data[t]
return smoothed
# 绘制根据不同选择绘制的图表（使用Seaborn）
def plot_tensorboard_data(all_seed_data_dict, tags, plot_type='mean_std', save_path=None, names=None, tag_names=None, num_train_steps_per_itr=2000):
sns.set_theme(style="whitegrid", font_scale=2.0) # 设置Seaborn的样式
if len(tags) == 6:
fig, axes = plt.subplots(2, 3, figsize=(15, 10))
axes = axes.flatten()
elif len(tags) == 4:
fig, axes = plt.subplots(1, 4, figsize=(18, 8))
axes = axes.flatten()
elif len(tags) == 2:
fig, axes = plt.subplots(1, 2, figsize=(24, 8))
axes = axes.flatten()
elif len(tags) == 1:
fig, axes = plt.subplots(1, 1, figsize=(12, 8))
axes = [axes]
# fig.suptitle(save_path.split('.')[0])
# 每个 base_dir 对应一个颜色
color_palette = sns.color_palette("tab10", len(all_seed_data_dict)) # 可选 "Set2" "tab10", "hsv", 或其他高对比度调色板
for i, tag in enumerate(tags):
for idx, (base_dir, all_seed_data) in enumerate(all_seed_data_dict.items()):
# 对不同的seed数据进行对齐
aligned_data = align_data(all_seed_data, tag)
# 计算均值和方差
mean, std = calculate_mean_and_std(aligned_data)
mean = np.squeeze(mean)
std = np.squeeze(std)
smoothed_mean = smooth_data(mean)
# 使用Seaborn绘制带误差条的曲线图
sns.lineplot(x=np.arange(len(smoothed_mean)) * (num_train_steps_per_itr / 1000), y=smoothed_mean, ax=axes[i], label=names[idx], color=color_palette[idx])
axes[i].fill_between(np.arange(len(smoothed_mean)) * (num_train_steps_per_itr / 1000), smoothed_mean - std, smoothed_mean + std, color=color_palette[idx], alpha=0.1)
axes[i].set_title(f'{tag_names[i]}', fontsize=30)
axes[i].set_xlabel(f'{tag_names[i].split(" ")[-1]} Steps(1e3)')
axes[i].set_ylabel('Average Return')
axes[i].grid(True, which='both', linestyle='--', linewidth=0.5, alpha=0.5)
axes[i].legend(loc='lower right', bbox_to_anchor=(1.0, 0.0), frameon=False)
plt.tight_layout()
# 如果提供了保存路径，则保存图片
if save_path:
plt.savefig(save_path)
print(f"图像已保存到：{save_path}")
# 用于执行绘图的接口
def visualize_tensorboard(base_dirs, tags=['tag1', 'tag2'], plot_type='mean_std', save_path=None, names=None, tags_names=None, num_train_steps_per_itr=2000):
all_seed_data_dict = {}
# 为每个 base_dir 提取数据
for base_dir in base_dirs:
all_seed_data = extract_seed_data(base_dir, tags)
all_seed_data_dict[base_dir] = all_seed_data
plot_tensorboard_data(all_seed_data_dict, tags, plot_type, save_path, names, tag_names=tags_names, num_train_steps_per_itr=num_train_steps_per_itr)
if __name__ == '__main__':
# base_dirs = ['logs/point-robot/FOCAL0135/focal_mix_baseline', 'logs/point-robot/FOCAL0135/focal_mix_z0_hvar_p10_weighted',
# 'logs/point-robot/CLASSIFIER0349/classifier_mix_baseline', 'logs/point-robot/CLASSIFIER0349/classifier_mix_z0_hvar_p10_weighted',
# 'logs/point-robot/UNICORN1237/unicorn_mix_baseline', 'logs/point-robot/UNICORN1237/unicorn_mix_z0_hvar_weighted',
# 'logs/point-robot/FOCAL0135/idaq_mix_baseline']
# base_dirs = ['logs/ant-goal/FOCAL0356/focal_mix_baseline', 'logs/ant-goal/FOCAL0356/focal_mix_z0_hvar_p0.1_weighted',
# 'logs/ant-goal/CLASSIFIER0123/classifier_mix_baseline', 'logs/ant-goal/CLASSIFIER0123/classifier_mix_z0_hvar_p5_weighted',
# 'logs/ant-goal/UNICORN1234/unicorn_mix_baseline0.1', 'logs/ant-goal/UNICORN1234/unicorn_mix_z0_p2_hvar0.1_weighted',
# 'logs/ant-goal/FOCAL0356/idaq_mix_baseline_-650']
base_dirs = ['logs/cheetah-vel/FOCAL04710/focal_mix_baseline', 'logs/cheetah-vel/FOCAL04710/focal_mix_z0_hvar_p2_weighted',
'logs/cheetah-vel/CLASSIFIER_0256/classifier_mix_baseline', 'logs/cheetah-vel/CLASSIFIER_0256/classifier_mix_z0_hvar_p10_weighted',
'logs/cheetah-vel/UNICORN_0123/unicorn_mix_baseline0.5', 'logs/cheetah-vel/UNICORN_0123/unicorn_z0_p0.5_hvar_0.5_weighted',
'logs/cheetah-vel/FOCAL04710/idaq_mix_baseline']
# base_dirs = ['logs/walker-rand-params/FOCAL0145/focal_mix_baseline', 'logs/walker-rand-params/FOCAL0145/focal_mix_z0_hvar_p5_weighted',
# 'logs/walker-rand-params/CLASSIFIER0124/classifier_mix_baseline', 'logs/walker-rand-params/CLASSIFIER0124/classifier_mix_z0_hvar_p25_weighted',
# 'logs/walker-rand-params/UNICORN0123/unicorn_mix_baseline0.5', 'logs/walker-rand-params/UNICORN0123/unicorn_mix_z0_p2_hvar0.5_weighted',
# 'logs/walker-rand-params/FOCAL0145/idaq_mix_baseline_r80'
# ]
# base_dirs = ['logs/hopper-rand-params/FOCAL0257/focal_mix_baseline', 'logs/hopper-rand-params/FOCAL0257/focal_mix_z0_hvar_p0.1_weighted',
# 'logs/hopper-rand-params/CLASSIFIER0124/classifier_mix_baseline', 'logs/hopper-rand-params/CLASSIFIER0124/classifier_mix_z0_hvar_p10_weighted',
# 'logs/hopper-rand-params/UNICORN037/unicorn_mix_baseline0.5', 'logs/hopper-rand-params/UNICORN037/unicorn_mix_z0_p2_hvar0.5_weighted',
# 'logs/hopper-rand-params/FOCAL0257/idaq_mix_baseline_r140'
# ]
names = ['focal', 'focal+ours', 'classifier', 'classifier+ours', 'unicorn', 'unicorn+ours', 'idaq']
plot_type = 'mean_std' # 可以选择 'mean_std', 'mean' 或 'std'
save_path = f'{base_dirs[0].split("/")[1]}'
num_train_steps_per_itr = 1000
visualize_tensorboard(base_dirs, ['Return/first_OnlineReturn_all_test_tasks'], plot_type, save_path+'1.pdf', names, ["First Episode Test"], num_train_steps_per_itr)
visualize_tensorboard(base_dirs, ['Return/second_OnlineReturn_all_test_tasks'], plot_type, save_path+'2.pdf', names, ["Second Episode Test"], num_train_steps_per_itr)
# visualize_tensorboard(base_dirs, ['Return/first_OnlineReturn_all_train_tasks'], plot_type, save_path+'_train1.pdf', names, ["First Episode Train"], num_train_steps_per_itr)
# visualize_tensorboard(base_dirs, ['Return/second_OnlineReturn_all_train_tasks'], plot_type, save_path+'_train2.pdf', names, ["Second Episode Train"], num_train_steps_per_itr)
# visualize_tensorboard(base_dirs, ['Return/Average_OfflineReturn_all_test_tasks'], plot_type, save_path+'_offline_test.pdf', names, ["Offline Return Test"], num_train_steps_per_itr)
# visualize_tensorboard(base_dirs, ['Return/Average_OfflineReturn_all_train_tasks'], plot_type, save_path+'_offline_train.pdf', names, ["Offline Return Train"], num_train_steps_per_itr)